Information processing apparatus, information processing method, pattern recognition apparatus, and pattern recognition method

Abstract

In a hierarchical neural network having a module structure, learning necessary for detection of a new feature class is executed by a processing module which has not finished learning yet and includes a plurality of neurons which should learn an unlearned feature class and have an undetermined receptor field structure by presenting a predetermined pattern to a data input layer. Thus, a feature class necessary for subject recognition can be learned automatically and efficiently.

Description

CROSS-REFERENCE TO RELATED APPLICATION[0001]This application is a continuation of application Ser. No. 11 / 149,699, filed Jun. 10, 2005, the entire disclosures of which are hereby incorporated by reference.FIELD OF THE INVENTION[0002]The present invention relates to a technique of executing arithmetic processing of a hierarchical neural network.BACKGROUND OF THE INVENTION[0003]As a method of enabling pattern recognition of a new category, a method using a neural network such as a multilayer perceptron or radial basis function network is known well. Especially, of models capable of reducing the influence of noise, a method of executing hierarchical template matching of local features (e.g., non-patent reference 1 (Yanagawa, Fukushima, and Yoshida, “Additional Learnable Neocognitron”, Technical Report of IEICE, NC2001-176, pp. 63-70, 2002)) is one of models capable of additional learning of a new category.[0004]In patent reference 1 (Japanese Patent Registration No. 2780454), the thres...

AI Technical Summary

Problems solved by technology

In the above-described prior arts, it is difficult to efficiently execute learning (adjustment of internal parameters such as the number of modules of the neural network and the connection weight distribution) for recognition of an arbitrary new pattern (unlearned feature pattern).

In addition, the type of a feature useful for internal representation of an arbitrary pattern cannot be known in advance.

However, it is not guaranteed that the feature class learned at this time is a new and effective feature category.

For this reason, it is difficult to execute efficient learning as a whole in recognizing a complex pattern.

However, since the reference pattern itself is input as the new category, it is not guaranteed that proper recognition can be done even when the pattern is deformed.

For this reason, it is difficult to execute efficient recognition and learning operation using a few hardware resources (circuit elements or memory space).

Hence, it is difficult to sufficiently implement adaptation to environment and versatility.

Images